The Study And Application Of Information Extraction Technology Based On High-Resolution Image

With the continuous competition in the field of the aerospace and the successful launching of the satellites,the resolution of the aviation data we obtain continues to improve,and the high-resolution images have been coming.The high-resolution data gives us a wealth of data at the same time,but brings a series of technical problems.The traditional information acquisition method is clearly unable to meet our needs for the high-resolution data.Because the information contained in high-resolution data is too rich,how can we quickly obtain the information we need from the complex data is a technical problem in the remote sensing field.In this paper,we take the GF-2 image data of Gong'an County in Hubei Province as an example,and choose a representative area having abundant objects features as the experimental area.As the image is divided into the urban and rural areas,we selected two typical areas to take the experiment.This paper elaborates the whole process of high resolution image information extraction: the image preprocessing,the method of the traditional image interpretation,the image segmentation,the common image feature and the image classification.This paper focuses on the object-oriented method of the interpretation of the high-resolution image.In order to demonstrate the obvious advantages of the object-oriented method,we set up the traditional interpretation method as a contrast,and compare these results.The results are as follows:(1)The accuracy of classification is 69.98% and 73.53% respectively.The accuracy of unsupervised classification is 34.41% and 46.02% respectively.There is "salt and pepper" phenomenon in the high resolution images,which makes their accuracy is not high.Because the unsupervised classification does not has the training samples,its accuracy is very lower.(2)The optimal segmentation scale of different objects is obtained by studying the standard deviation of the mean value of the brightness at different scales.When the standard deviation of the mean value of the brightness is biggest,the separation scale of the object is the best.(3)The accuracy of the object-oriented classification is 76.44% and 88.68% respectively.The accuracy of the object-oriented classification has a certain improvement.Because the first experimental area has complex types of buildings,and the houses and the roads have the similar spectral information,and the water and the shadow of a variety of features have the similar spectral information,it makes the classification accuracy is not ideal.